A government-enterprise service platform method, system, device and storage medium constructed by an artificial intelligence algorithm

Abstract

This invention provides a method, system, device, and storage medium for a government-enterprise service platform constructed using artificial intelligence algorithms, relating to the field of computer information technology. It utilizes computer vision, natural language processing, and multimodal information fusion technologies to construct enterprise profiles, analyze external demands, and perform information matching and recommendation. The method for constructing the government-enterprise integrated service platform of this invention specifically includes: acquiring enterprise video data and enterprise text data; processing the enterprise video data to generate video summary information; parsing the enterprise text data to generate an enterprise knowledge base representing the enterprise's business priorities; expanding the video summary information using the enterprise knowledge base to generate a feature containing at least one production information feature; acquiring external information data and extracting demand information features from the external information data; and pushing targeted information based on the correlation analysis between the production information features and the demand information features. Through the above steps, this invention achieves accurate two-way profiling and matching of an enterprise's internal production capabilities and external market information.

Description

Technical Field

[0001] This application relates to the field of computer information technology, and more particularly to a method, system, device, and storage medium for a government and enterprise service platform constructed using artificial intelligence algorithms. More specifically, this application relates to utilizing computer vision, natural language processing, and multimodal information fusion technologies to construct enterprise profiles, analyze external needs, and perform information matching and recommendation. Background Technology

[0002] In the digital age, the internet, as the primary carrier of information dissemination, has reached unprecedented levels in breadth, depth, and speed. Social media, news portals, industry forums, government websites, and other information sources are generating and disseminating massive amounts of data at an unprecedented rate. While this explosive growth of information brings convenience to society, it has also given rise to a severe problem of information overload. For businesses participating in the market, this information overload constitutes a hidden and enormous operational challenge. Not all massive amounts of information have equal value; much of it is irrelevant and redundant. If businesses want to accurately and efficiently extract information closely related to their production and operations and of decision-making value from this vast ocean of information, they need to invest significant human, material, and time resources. Traditional information acquisition models, relying on manual screening and experience-based judgment, are proving inadequate under the onslaught of information, and their inefficiencies, narrow coverage, and poor timeliness are becoming increasingly apparent. This often leads to companies being slow to react to the rapidly changing market environment and failing to seize fleeting business opportunities, such as new market demands, emerging technological trends, potential risks in the supply chain, or new policy guidelines. As a result, they miss out on development opportunities in the fierce market competition and may even make wrong strategic decisions, affecting the company's sustainable and healthy development.

[0003] This problem of information asymmetry and low processing efficiency has created an information silo dilemma involving producers, consumers, and management service providers in various links of the industrial chain, which seriously restricts the optimal allocation of market resources and overall operational efficiency.

[0004] For manufacturing enterprises, the core dilemma lies in the disconnect between production and the market. Production decisions, such as expanding or reducing production capacity for a particular product, adjusting production plans, undertaking technological upgrades, or developing new products, all require accurate predictions of external market demand. However, in an environment of information overload, enterprises struggle to systematically and proactively glean market signals that align with their specific products and production capacity from the vast amounts of fragmented news reports, industry analyses, and social media discussions. For example, a news article about a region's new energy vehicle subsidy policy might be a significant boon for a company producing related auto parts, but this information is likely to be lost in the sea of ​​daily news, preventing the company from adjusting production in time to cope with potential order growth. Enterprises lack an effective technological means to automatically and accurately link their own internal, dynamic production capabilities with external, unstructured market demand information. Consequently, production decisions are often based on lagging historical sales data or vague macroeconomic judgments, lacking foresight and specificity.

[0005] For those who demand products or services, such as downstream manufacturers, buyers, or large project owners, the dilemma they face is the difficulty in finding and evaluating suppliers. When a demander needs to find suppliers capable of producing specific specifications, quality standards, and with corresponding production capacity, they are similarly caught in an information fog. Although there are numerous business directories and B2B platforms on the internet, the information on these platforms is mostly self-reported by companies, static, and often embellished promotional content. Demanders find it difficult to ascertain a company's true and real-time production status.

[0006] For government departments or related service organizations aiming to serve and guide enterprise development, various policy documents, industry matchmaking events, and industry risk warnings are issued to promote local economic development and industrial upgrading. However, accurately delivering these valuable service resources to the specific enterprises that need them most and meet the criteria has always been a major challenge. Traditional, broad-based policy promotion methods have extremely low conversion rates. This is because they lack the ability to perceive the micro-level and dynamic operational status of enterprises, making it impossible to accurately determine which enterprises are undergoing technological upgrades, which have potential supply chain risks requiring risk warnings, or which enterprises' products happen to meet the needs of a particular emerging market and require market information and channel support. This "blind men and the elephant" service model results in a large amount of information not being accurately delivered to enterprises, and enterprises not being able to fully enjoy timely information.

[0007] In view of the technical problems existing in the aforementioned background technologies, such as the difficulty for enterprises to obtain effective decision-making basis from massive amounts of information, this application proposes a method and system for a government-enterprise service platform built through artificial intelligence algorithms. By deeply integrating computer vision and natural language processing technologies, it achieves accurate two-way profiling and matching of an enterprise's internal production capabilities and external market information. This helps manufacturing enterprises proactively discover market opportunities, provides accurate target enterprise profiles, and effectively breaks down information silos. Summary of the Invention

[0008] This invention provides a method for building a government-enterprise service platform using artificial intelligence algorithms, the method specifically including the following steps:

[0009] Acquire enterprise video data and enterprise text data;

[0010] The enterprise video data is processed to generate video summary information;

[0011] The enterprise text data is parsed to generate an enterprise knowledge base that represents the enterprise's business priorities;

[0012] The video summary information is expanded using the enterprise knowledge base to generate a feature containing at least one production information feature;

[0013] Acquire external information data and extract demand information features from the external information data;

[0014] Based on the correlation analysis between the production information features and the demand information features, the target push information is pushed.

[0015] This specification also proposes a government-enterprise service platform system built using artificial intelligence algorithms, which includes:

[0016] Acquisition Module: The acquisition module obtains enterprise video data and enterprise text data;

[0017] Summary information generation module: The summary information generation module processes the enterprise video data to generate video summary information;

[0018] Summary information expansion module: The summary information expansion module uses the enterprise knowledge base to expand the video summary information and generate a feature containing at least one production information feature;

[0019] Demand information feature extraction module: Expands the video summary information using the enterprise knowledge base to generate features containing at least one production information feature;

[0020] The push module pushes the target push information based on the correlation analysis between the production information features and the demand information features.

[0021] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the aforementioned method for building a government and enterprise service platform using artificial intelligence algorithms.

[0022] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the aforementioned method for building a government and enterprise service platform using an artificial intelligence algorithm.

[0023] The present invention proposes a method and system for building a government-enterprise service platform using artificial intelligence algorithms, which aims to overcome the information silos caused by information overload, difficulties in matching market supply and demand, and poor communication channels between government and enterprises in the existing technologies.

[0024] First, the enterprise profiling construction process based on multimodal data fusion in this invention ensures the objectivity and dynamism of the enterprise profile. Existing technologies generally use a single text information source, which cannot guarantee the authenticity and timeliness of its content. This invention uses video data of the enterprise's actual production and operation as the initial data source for profiling, directly obtaining quantifiable and verifiable objective indicators about the enterprise's physical world operations. Based on this, it further supplements and expands the objective production facts derived from video analysis by analyzing enterprise text information such as patents and qualifications, adding dimensions such as technical capabilities and industry affiliation, thereby improving the accuracy of information delivery.

[0025] Secondly, in the step of processing enterprise video data to generate video summary information, this invention employs an adaptive method, solving the technical challenges of general applications. Facing the challenges of diverse products and varying production line layouts across different enterprises, the two-stage, adaptive product recognition method proposed in this invention combines foreground object segmentation with few-sample comparison, achieving accurate identification of in-production products without needing to retrain the model for each new product, demonstrating strong versatility and scalability. Simultaneously, its virtual detection line automatic calibration method based on motion pattern analysis can autonomously determine the optimal counting position according to the actual object flow patterns in the video, avoiding the limitations of requiring manual intervention to set the analysis area, significantly improving the system's automation level and robustness in complex scenarios. The application of these adaptive technologies ensures that the platform can provide standardized, quantifiable production efficiency and human resource input pattern analysis for a massive number of enterprises in a low-cost, high-efficiency manner.

[0026] In summary, this invention, through a multimodal profiling framework and adaptive video analytics, can generate an objective and dynamically updated production capacity profile for enterprises. Based on this high-quality profile, the platform's matching and delivery of external information is no longer based on simple keyword matching, but rather on a comprehensive multi-dimensional assessment of the enterprise's production capacity, technological capabilities, and industry positioning. Therefore, this invention effectively solves the problem of information overload faced by enterprises, enabling them to obtain timely and crucial information that has substantial decision-making value for their production and operations. Attached Figure Description

[0027] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0028] Figure 1 This is a flowchart of the government and enterprise service platform constructed using artificial intelligence algorithms according to the present invention. Detailed Implementation

[0029] The embodiments of this application will now be described in detail with reference to the accompanying drawings.

[0030] The following specific examples illustrate the implementation of this application. Those skilled in the art can easily understand other advantages and effects of this application from the content disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. This application can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of this application. It should be noted that, in the absence of conflict, the following embodiments and features in the embodiments can be combined with each other. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0031] It should be noted that various aspects of embodiments within the scope of the appended claims are described below. It will be apparent that the aspects described herein can be embodied in a wide variety of forms, and any particular structure and / or function described herein is merely illustrative. Based on this application, those skilled in the art will understand that one aspect described herein can be implemented independently of any other aspect, and two or more of these aspects can be combined in various ways. For example, any number and aspects set forth herein can be used to implement the device and / or practice the method. Additionally, this device and / or method can be implemented using structures and / or functionalities other than one or more of the aspects set forth herein.

[0032] Additionally, specific details are provided in the following description to facilitate a thorough understanding of the examples. However, those skilled in the art will understand that practice can be carried out without these specific details.

[0033] This specification presents an embodiment of a method for building a government-enterprise service platform using artificial intelligence algorithms. The method specifically includes the following steps:

[0034] Acquire enterprise video data and enterprise text data;

[0035] The enterprise video data is processed to generate video summary information;

[0036] The enterprise text data is parsed to generate an enterprise knowledge base that represents the enterprise's business priorities;

[0037] The video summary information is expanded using the enterprise knowledge base to generate a feature containing at least one production information feature;

[0038] Acquire external information data and extract demand information features from the external information data;

[0039] Based on the correlation analysis between the production information features and the demand information features, the target push information is pushed.

[0040] To make the implementation of this application clearer, the implementation of this application will be described in detail below with reference to specific embodiments. In the following embodiments, the government and enterprise integrated service platform used to implement the method of this application can be deployed in a cluster of one or more servers and communicate with enterprise clients through a network.

[0041] This application provides a method for building a government-enterprise service platform using artificial intelligence algorithms. The initial step involves acquiring video and text data from the target enterprise. In one specific embodiment, this acquisition process begins with an authorized representative of the target enterprise creating an account and completing real-name authentication on the government-enterprise integrated service platform. During account creation, the platform's user interface presents the authorized representative with a clear and detailed "Data Processing and Service Authorization Agreement." This agreement explicitly lists the types of data required by the platform to provide subsequent services, the acquisition methods, the purpose of use, and data security measures. The authorized representative must agree to all terms of the agreement through interactive actions such as actively checking a confirmation box, thus providing a clear and traceable legal basis for all subsequent data collection activities.

[0042] After obtaining explicit authorization from the user, the platform's data acquisition module is activated and begins acquiring enterprise video data. This embodiment provides two complementary video data acquisition methods, which enterprises can choose according to their own circumstances. The first method is real-time video stream access, suitable for continuous and dynamic production and operation analysis. The authorized representative enters the Real-Time Streaming Protocol (RTSP) address and access credentials of existing security cameras deployed in non-confidential production and operation areas (such as product packaging areas and warehouse entry / exit points) through the video source management interface provided by the platform. At the same time, the authorized representative must label each video stream with its corresponding scene context, such as selecting from preset labels such as final assembly line or logistics loading and unloading area, and setting the effective working period for platform analysis. The second method is discrete video file upload, suitable for retrospective analysis of specific batches or events. The authorized representative selects locally stored video files that record production and operation activities through the platform's upload interface, and must supplement them with key metadata during upload, including the video recording date and its corresponding scene context. Through these two methods, the platform can obtain video data with clear scene context information that truly reflects the enterprise's operational status.

[0043] Simultaneously, the platform acquires enterprise text data. This acquisition process is also conducted with the user's explicit authorization. The platform traverses and extracts content from the publicly accessible pages of the website using the domain name of the enterprise's official website provided during registration, completely capturing all text content and recording product image information. Simultaneously, the platform calls upon a public data interface service. This service uses the certified enterprise's full name as the query keyword and securely connects to a public database via an application programming interface to query and retrieve all publicly available patents, trademarks, and software copyrights under the enterprise's name, especially key texts such as the invention title, abstract, and claims of patents. Furthermore, the platform provides a qualification certification management interface, allowing authorized representatives to independently upload digital copies of their enterprise's patents, trademarks, software copyrights, and various industry qualification certificates and management system certification certificates, such as high-tech enterprise certificates and ISO9001 quality management system certification certificates.

[0044] All raw data acquired through the aforementioned channels, whether video streams, web page text, patent abstracts, or qualification certificate images, undergoes a unified preprocessing and structured storage step before entering the subsequent analysis process. The system immediately attaches a set of unified metadata tags to each received data item. The most crucial tags include: a unique enterprise identifier (EnterpriseID) to associate with a specific company, a tag indicating the data source type (RTSP_STREAM, WEBSITE_CRAWL, PATENT_API, ISO_UPLOAD), and a data acquisition timestamp accurate to the second. After metadata annotation, this heterogeneous data is stored in the platform's distributed multimodal database. Video data is stored in a time-series database or a distributed file system, while structured or semi-structured data such as text and qualification information are stored in a relational database.

[0045] In one specific embodiment of this application, after the enterprise video data and enterprise text data, the video analysis module of the government and enterprise integrated service platform will perform the operation of processing the enterprise video data to generate video summary information.

[0046] The step of processing the enterprise video data to generate video summary information specifically includes:

[0047] Identify the identifier of products in production Calculate the production efficiency corresponding to the identifier of the product in production, and obtain the production efficiency category based on a preset efficiency comparison. The average number of workers is estimated based on the attitude estimation algorithm, and the labor input rate category is obtained by comparing it with a preset labor threshold. The video summary information is obtained by integration. .

[0048] The video analysis module retrieves video data from a multimodal database and decodes it into a continuous sequence of image frames at a specific frame rate. For each frame... ,in Representing timestamps, the system performs product identification. The platform needs to handle thousands of different products from various companies, making it impossible to train a universal detection model encompassing all products for accurate identification of each. To address this issue, this invention employs a two-stage, adaptive identification method. This method relies solely on the company product image library acquired during the data acquisition phase. The image data in this library is obtained from the company's official website or uploaded by the user.

[0049] The first stage of the two-stage, adaptive recognition method involves segmenting foreground objects unrelated to the product. This invention does not directly identify the product; instead, it first determines the area of ​​the processed product being moved within the video frame. By applying a background subtraction method based on a Gaussian mixture model (GMM), the system establishes a dynamically updated background model for the video scene. For each input frame image By comparing it with the background model By performing difference operations and morphological processing, the system can segment the foreground region, which consists of all moving or changing objects in the image. These foreground regions are further clustered and bounded, generating a bounding box for each individual moving object. .in, These represent the center coordinates, width, and height of the bounding box of the i-th target to be identified, respectively. The output of this stage is the precise location of all targets to be identified in the image. It does not rely on any prior product knowledge and has extremely high versatility.

[0050] The second stage of the two-stage, adaptive recognition method is feature comparison recognition based on few-shot learning. In this stage, the system determines which product the target segmented in the previous step is. The system pre-loads a deep neural network with powerful feature extraction capabilities, pre-trained on a large public image dataset. Preferably, a CNN or ResNet backbone network is used. Its function is not classification, but rather as a general feature extractor. It utilizes a backbone network. , to store the enterprise's product image library

[0051] Each standard product image is input into the backbone network, and its corresponding baseline features are extracted and stored. ,in - These are product identifiers, - These are the standard images corresponding to each product identifier, and then, for each video frame... Each bounding box obtained through foreground segmentation The system crops out the image region within the bounding box and inputs it into the feature extractor. In this process, its observed feature vector is obtained. By calculating the observed feature vector With all benchmark feature vectors in the library similarity To perform matching:

[0052] in Represents the dot product of vectors. The Euclidean norm of the vectors is represented. The system selects vectors with the highest similarity exceeding a preset threshold. The matching result is used as the identification candidate for the object.

[0053] By voting on the identification results of the same tracked object in multiple consecutive frames, a unique and most reliable in-production product identifier is ultimately determined for the entire video. .

[0054] The product identifiers in production in the video were determined using a two-stage method. Next, the video analytics module needs to quantify the product's production efficiency. To adapt to the diverse production layouts of enterprises, this invention employs an automatic virtual inspection line calibration method based on motion pattern analysis. The core idea of ​​this method is that the system learns the overall motion patterns of objects in the video, adaptively identifies the most stable and consistent flow path of the product, and sets the inspection line at the key nodes of that path.

[0055] In a preferred embodiment, the system handles two consecutive frames of images. and Calculate an optical flow field , where vector Indicates the location The pixels from Frame to The motion displacement of the frame. By averaging the optical flow field over time within a statistical period, a stable optical flow map characterizing the average motion pattern in the scene can be obtained. .

[0056] The system stabilizes optical flow maps The system analyzes the data to identify the main motion path. It divides the entire screen into a... The grid. For each grid cell... Calculate the average motion vector of all pixels within it. For those with large motion amplitude (i.e. ,in Adjacent grid cells with relatively consistent orientations and a motion saliency threshold are defined as high-activity regions. Clustering is used to find the highest-activity region and this region is defined as the main transport channel for this scene, denoted as […]. .

[0057] After determining the main transport channel Then, calculate the weighted average motion vector of all pixels in the region. The direction of this vector is the main direction of travel of the product in this channel.

[0058]

[0059] Characterizing motion vectors The direction vector, weighted average with Motion Amplitude As a weight, pixels moving at high speeds contribute more to the calculation of the main direction.

[0060] testing line The optimal position should be perpendicular to the main direction of travel. And it runs across the main transport channel The center. Definition The geometric center point is Due to the testing line normal vector With the main direction of travel Parallel, that is Then the detection line The equation of the straight line can be determined as:

[0061]

[0062] For each tracked product, continuously monitor the coordinates of its bounding box center point. By determining whether the center point crosses the detection line. The coordinates of the center point of the bounding box at time t. Substitute the coordinates of the bounding box center point at migration time t-1 into the detection line. In the linear equation, if the product of the results is negative, it is determined that the product has crossed the detection line L, and product counting is performed. This is done within a preset statistical time period. Inside, products The total count value is Then the production efficiency of this product. The calculation formula is: ,based on By comparing with the preset efficiency, the production efficiency category is obtained. The production efficiency category It is defined as three categories: high, medium, and low.

[0063] The video analytics module also needs to estimate human resource input. The human body analysis model in this embodiment... A mature multi-user pose estimation algorithm architecture is adopted. In a preferred embodiment, This model implements human pose recognition based on the OpenPose network. OpenPose uses a convolutional neural network to extract feature maps from the entire image. Then, through two parallel branches, it predicts the confidence maps and connectivity relationships (PAFs) between keypoints of all body parts (such as the head, neck, shoulders, elbows, and wrists, totaling 18 keypoints). Using PAFs, keypoints belonging to different people are correctly grouped, thus constructing an independent human pose skeleton in the image. For each input frame Output a set containing q skeletons. Each skeleton represents a detected person. Therefore, in time... In the scenario, the actual number of workers That is, a set The number of skeletons, q. To obtain the statistical period. The system calculates the average number of people within a given period as a stable indicator. :

[0064]

[0065] in It represents the total number of frames processed.

[0066] based on Compared with the preset manpower threshold, the manpower input rate category is obtained. The category of human resource input rate It is defined as three categories: labor-intensive, personnel-assisted, and automated industries.

[0067] The video intelligent analysis module integrates the outputs of the above three tasks into structured video summary information. :

[0068] .

[0069] Existing technologies for assessing enterprise production capacity rely on subjective reporting by the enterprise or inefficient manual due diligence. This invention, through video analysis, directly extracts in-production products from images of the production site, calculates production efficiency, and estimates labor input—all objective production data. It makes the enterprise's production workshop transparent and data-driven, providing an objective foundation for all subsequent analysis and matching, avoiding information distortion problems. Furthermore, it's worth noting that this method achieves a high degree of adaptability. Faced with the complex reality of diverse enterprise products and varying production layouts, this invention does not employ the traditional, rigid method that requires training models separately for each product. It uses a two-stage, adaptive recognition method, combining foreground segmentation with few-sample learning feature comparison, to achieve rapid and accurate identification of any product from any enterprise, exhibiting strong versatility and scalability. The virtual detection line automatic calibration method based on motion pattern analysis can adapt to any production line layout, while quantifying capacity and labor costs. This enables subsequent accurate matching and industry benchmark analysis.

[0070] Parsing the enterprise text data to generate an enterprise knowledge base includes constructing industry knowledge entries:

[0071] It calls a pre-built authentication standard rule library, where each rule corresponds to an authentication standard and defines the inference type as either direct inference or semantic inference;

[0072] The uploaded qualification certificates are extracted to identify the standard name of the certificate and the description text of the certification scope; and the certification standard rule base is queried based on the standard name, and industry knowledge entries are constructed by direct industry mapping or keyword extraction from the description text of the certification scope according to the inference type found in the query.

[0073] In one specific embodiment of the present invention, to achieve the parsing of industry and compliance information in enterprise text data, the platform's text information parsing module pre-builds a structured authentication standard rule base in the offline state of the platform. Each rule in the base corresponds to a specific authentication standard and defines how to parse that standard. Specifically, each rule contains three core fields: 1) Standard name, i.e., the official name on the certification certificate, such as "IATF 16949:2016", which serves as the primary key for the query; 2) Inference type, whose value is "direct inference" or "semantic inference", used to indicate the subsequent processing logic; 3) Direct industry mapping, this field is only valid when the "inference type" is "direct inference", directly giving the industry category corresponding to the standard, such as "automobile manufacturing".

[0074] Specifically, an exemplary certification standard rule base includes: one rule is {Standard Name: "IATF 16949:2016", Inference Type: "Direct Inference", Direct Industry Mapping: "Automotive Manufacturing"}; another rule is {Standard Name: "ISO9001:2015", Inference Type: "Semantic Inference", Direct Industry Mapping: NULL}.

[0075] The reason for distinguishing between direct inference and semantic inference in standards lies in the fact that industry-specific qualifications are established specifically for a particular industry, and obtaining such a certificate is equivalent to declaring the company's industry affiliation. For example, IATF 16949: This is the quality management system standard for the automotive industry. Any company that obtains this certification must have a business closely related to the automotive supply chain. Therefore, IATF 16949 -> Automotive Manufacturing is a very clear mapping rule. Another example is AS9100: This is the quality management system requirement for the aerospace and defense organizations. Therefore, AS9100 -> Aerospace Industry.

[0076] For general certifications (such as ISO 9001 quality management system), the certification certificate will clearly indicate the scope of certification. This scope describes which specific products or services the quality system covers in the company's production process.

[0077] For example, a company holding an ISO 9001 certificate does not inherently indicate its industry. However, the scope of its certification might state: "Design, manufacture, and sale of server racks for data centers." This scope allows the company to be mapped to industries such as electronics manufacturing or IT infrastructure.

[0078] After the certification standard rule base is built and deployed on the platform, when a specific enterprise user uploads a digital copy of its qualification certificate on the platform, the text information parsing module calls the rule base to perform real-time parsing and association operations to generate the enterprise's industry knowledge entry KC and store it in its exclusive enterprise knowledge base BKB.

[0079] The parsing process first activates the optical character recognition module to extract text from the uploaded certificate image, converting its content into a machine-readable string. Then, it identifies and extracts key information from the string: the certificate's standard name and a descriptive text describing the scope of authentication.

[0080] The system uses the standard name of the identified certificate as the query condition to search the certification standard rule base. If the rule found has a direct inference type, the system directly obtains the value of its direct industry mapping field (e.g., "Automotive Manufacturing") and uses it to construct the industry knowledge entry KC. Conversely, if the rule found has a semantic inference type (e.g., matching "ISO 9001:2015"), further processing of the "certification scope description text" is required. In this case, the system extracts keywords from the previously extracted certification scope description text and uses these keywords to construct the industry knowledge entry KC.

[0081] This method improves the accuracy and flexibility of industry positioning by distinguishing between direct inference and semantic inference. For industry-specific qualifications like IATF 16949, the system can provide a definitive, high-confidence industry classification; for universal qualifications like ISO 9001, the system can perform semantic analysis based on the description of its certification scope to infer its specific industry application in a particular business scenario, avoiding the drawback of simply categorizing all companies. Furthermore, the industry knowledge entry KC generated by this method is a structured information entity, not just a label.

[0082] Parsing the enterprise text data to generate an enterprise knowledge base also includes constructing technical knowledge entries:

[0083] Call the patent classification database to convert the IPC or CPC classification numbers of patent documents into text descriptions of the classification numbers;

[0084] The invention title, abstract, and classification number text description of the patent are concatenated into a text corpus; and a set of high-confidence technical keywords is selected by calculating the technical relevance score or TF-IDF weight score of candidate keywords in the text corpus in order to construct technical knowledge entries.

[0085] In another specific embodiment of the present invention, the platform's text information parsing module then parses the patent documents. The goal of this process is to create a technical knowledge entry, denoted as KP, for each patent document and import it into the enterprise's knowledge base (BKB). In one specific embodiment, the parsing process includes two processing branches: one is the parsing of structured data such as IPC / CPC classification numbers, and the other is the parsing of unstructured text such as invention titles and abstracts.

[0086] The parsing process first processes the highly structured International Patent Classification (IPC) and Collaborative Patent Classification (CPC) numbers in the patent documents. These classification numbers are globally recognized standardized codes used to identify the technical fields of inventions. The text information parsing module incorporates a patent classification database pre-loaded during platform construction. This database fully stores the hierarchical classification system of IPC and CPC, as well as the official detailed technical explanation text corresponding to each classification number. When the system obtains a patent (e.g., patent number CN123456B) and its accompanying list of classification numbers, such as 'H02P 21 / 00', 'G05B 19 / 40', the system queries the patent classification database. For each classification number, the query operation returns a complete text description of the classification number, from the top-level classification to the bottom-level classification. For example, for H02P 21 / 00, the system will retrieve and generate a structured technical field description, such as {'Classification System':'IPC', 'Path':'H Department Electrical Engineering > H02 Chapter Power Generation, Transformation or Distribution > H02P Group Motor Control > 21 / 00 Group Vector Control'}. Through this step, the system transforms the classification number into a text description with a clear hierarchy and explicit technical meaning, which can be used for subsequent semantic analysis, providing an objective field positioning for this patented technology.

[0087] The parsing module extracts unstructured text content such as the invention title and abstract from patent documents. The system then concatenates the invention title, abstract, and classification number into a single text corpus. Subsequently, a pre-trained Transformer-based language model was invoked to compute the corpus. Each candidate keyword Technical relevance score This score is used to identify words in a text that not only appear frequently but are also semantically central. The calculation formula is as follows:

[0088]

[0089] in, Keywords In the text The frequency of occurrence in S is used to reflect its fundamental importance; S and These are the total number of layers in the Transformer model and the number of attention heads per layer, respectively. This represents the s-th layer of the model. In the attention head, text All other words within the keyword The sum of attention weights. This accumulated attention weight value is used to measure the keyword's... It plays a central role in the deep semantic structure of the entire text. The system will filter out all relevance scores. Exceeding the preset threshold The keywords form a set of highly reliable technical keywords.

[0090] When the platform's computing resources are limited, this invention provides an alternative embodiment, which differs in its method of processing unstructured text such as the invention title and abstract. The system concatenates the invention title, abstract, and classification number text description into a single document to be analyzed. Subsequently, the system invokes the TF-IDF algorithm to identify and extract core technical keywords. To ensure the effectiveness of the IDF calculation, the platform pre-stores a domain-related background technology corpus during its construction, denoted as [missing information]. This corpus consists of publicly available patent documents related to the service areas of this platform. The execution process of the TF-IDF algorithm is as follows: Calculate candidate keywords. In the current document The term frequency (TF) in [the context of the text]. In a specific implementation, its calculation formula is:

[0091]

[0092] in, Keywords In the document The number of times it appears in It is a document The total number of words. This value reflects the number of keywords. Local importance in the current patent description.

[0093] System calculation keywords Inverse Document Frequency (IDF). This value reflects the keyword's inverse document frequency (IDF). The formula for calculating the prevalence or scarcity of something across the entire technology field is as follows:

[0094]

[0095] in, It is a background technology corpus The total number of documents in the document. The corpus contains keywords The number of documents. According to the IDF formula, a common term that appears frequently in all documents (such as "system" or "method") will have a low IDF value, while a technical term that appears only in a few patents in specific technical fields (such as "annealing algorithm" or "hash collision") will have a high IDF value.

[0096] The system multiplies the TF value and IDF value to obtain the keyword. In the document The final TF-IDF weight score :

[0097]

[0098] Weighted Score It combines the local importance of keywords in the current document with their global scarcity across the entire technical field. The system analyzes the document... After word segmentation and stop word removal, the TF-IDF score of each remaining word is calculated, and all words with scores higher than a preset threshold are selected. The vocabulary forms a set of highly reliable technical keywords for this patent.

[0099] Each technical knowledge entry (KP) records the following information in a structured form: a patent identifier, i.e., the application number or publication number of the patent, used for unique identification and traceability; an invention title, which provides a description of the technology; and a set of technical keywords, which contains all the core words that can accurately describe its technological innovation and are selected through relevance scores. This complete technical knowledge entry (KP) is then stored in the target company's enterprise knowledge base (BKB).

[0100] The steps of expanding the video summary information using the enterprise knowledge base specifically include: semantically vectorizing the in-production product identifiers in the video summary information to generate product identifier feature vectors; traversing the industry knowledge entries in the enterprise knowledge base, calculating the similarity between the product identifier feature vectors and vectors representing industry semantics, and using the industry semantic vector with the highest similarity as the expanded information; and traversing the technical knowledge entries in the enterprise knowledge base, calculating the similarity between the product identifier feature vectors and vectors representing technical semantics, and using the technical semantic vector with the highest similarity as the expanded information.

[0101] In an embodiment of the present invention, after the enterprise knowledge base (BKB) is constructed, the platform utilizes the enterprise knowledge base to analyze video summary information. To expand the identifier for products in production. Construct a production information feature that includes its industry affiliation and technological support. .

[0102] Enterprise Knowledge Base (BKB) includes A collection of industry knowledge items and A collection of technical knowledge items .

[0103] Identifiers for products in production Semantic vectorization is performed. To enable in-depth semantic comparison with industry and technical content in the knowledge base, a pre-trained Transformer model is invoked to vectorize the product identifier. As input, a product identifier feature vector is generated, defined as follows: .

[0104] The platform traverses every industry knowledge entry in the enterprise knowledge base (BKB). For each entry, the system constructs a semantic vector representing the industry. If the inference type for this entry is semantic inference, then keyword extraction of the authentication scope description text (such as "servo motor controller") is used to generate the result. If the inference type is direct inference, then its industry type (such as "automobile manufacturing") is used itself to generate the inference. Calculate the product identification feature vector. Industry semantic vectors The similarity between them. The industry semantic vectors with the highest similarity. Added as supplementary information to the video summary information middle.

[0105] At the same time, the platform traverses every technical knowledge entry (KP) in the enterprise knowledge base (BKB). For each KP, the platform iterates through every technical term in the set of technical keywords. For each word Using the same pre-trained Transformer model, generate its corresponding technical semantic vector. Product identification feature vector With each technical semantic vector The similarity between them will be used to determine the technical semantic vectors with the highest similarity. Added as supplementary information to the video summary information Production information characteristics are obtained from this.

[0106] This invention constructs a complete information chain of "industry-product-technology" through semantic expansion both "upward" and "downward." It identifies the industry affiliation and technological support for a specific product seen in a video, ensuring that the enterprise profile is not merely a simple accumulation of various information dimensions, but rather an organic whole with inherent logic and interrelationships. Ultimately, the generated production information features have each dimension cross-validated and enriched by multi-source information, resulting in an information range and credibility far exceeding that of any single-source information in existing technologies.

[0107] The correlation analysis based on the production information features and the demand information features specifically includes: extracting a set of keywords representing industry, product, technology, production efficiency demand, and human resource input demand from external information data; vectorizing the keyword sets of industry, product, and technology respectively, and comparing them with the corresponding vectors in the production information features; classifying the keyword sets of production efficiency demand and human resource input demand, and comparing them with the corresponding categories in the production information features; when at least four of the five dimensions of industry, product, technology, production efficiency, and human resource input are successfully matched, the target push information is determined.

[0108] This invention employs a multi-stage pipeline analysis method based on domain dictionaries and rule inference to extract features from external information. In this embodiment, the analysis method pre-builds a matching dictionary in an offline platform state. The construction process includes: 1) a product and service dictionary (Dict_prod), which aggregates publicly available industry product catalogs and product list data from B2B websites to form a dictionary containing numerous product and service names; 2) a technology and standards dictionary (Dict_tech), which constructs a technology field dictionary by parsing publicly available technical standard documents (ISO standards), professional technical encyclopedias, and detailed descriptions of patent classifications; and 3) an industry classification dictionary (Dict_ind), based on the Global Industry Classification Standard (GICS) and the National Economic Industry Classification (GB / T). 4754 Construct a hierarchical industry name dictionary; 4) Production efficiency demand dictionary Dict_Eff, including the category Dict_Eff_High (e.g., keywords "mass production", "emergency delivery", etc.), the category Dict_Eff_Medium (e.g., keywords "regular orders", "stable supply", etc.) and the category Dict_Eff_Low (e.g., keywords "small batch customization", "prototype trial production", etc.); 5) Human resource input demand dictionary Dict_Labor, including the category Dict_Labor_Intensive (e.g., keywords "handcrafting", "labor-intensive", etc.), the category Dict_Labor_Assisted (e.g., keywords "human-machine collaboration", "semi-automatic", etc.) and the category Dict_Labor_Automated (e.g., keywords "unmanned workshop", "fully automated", etc.).

[0109] When a new external information document is legally obtained At that time, after text preprocessing, the analysis pipeline uses a multi-modal matching algorithm to match and extract all keywords belonging to the matching dictionary in the text, forming five temporary keyword sets: Set_Dict_ind, Set_Dict_tech, Set_Dict_ind, Set_Dict_Eff, and Set_Dict_Labor.

[0110] For the three dimensions of industry, product, and technology that require vector comparison, the keywords arbitrarily extracted from each keyword set are vectorized and compared with their corresponding... , , If the similarity is higher than the preset threshold, the comparison of the corresponding dimension is considered successful.

[0111] For the two dimensions requiring categorical comparison—production efficiency and human resource input—a rule-based classification operation is performed, using the category with the most frequent keywords as the final category for that dimension, and then comparing it with... A category comparison is performed; if the categories match, the comparison of the corresponding dimension is considered successful.

[0112] Taking production efficiency as an example, the number of keywords in the text that matched the dictionaries Dict_Eff_High, Dict_Eff_Medium, and Dict_Eff_Low was counted. Then, the category corresponding to the dictionary with the highest number of matches was selected as the final production efficiency requirement category. For example, if there are 3 words in the "high" category and 1 word in the "medium" category, then... It was assigned the value 'high'.

[0113] When the comparison is successful in any four or more of the five dimensions of industry, product, technology, production efficiency and human resources input, the external information document will be recommended to the enterprise.

[0114] This specification also proposes a government-enterprise service platform system built using artificial intelligence algorithms, the device comprising:

[0115] Acquisition Module: The acquisition module obtains enterprise video data and enterprise text data;

[0116] Summary information generation module: The summary information generation module processes the enterprise video data to generate video summary information;

[0117] Summary information expansion module: The summary information expansion module uses the enterprise knowledge base to expand the video summary information and generate a feature containing at least one production information feature;

[0118] Demand information feature extraction module: Expands the video summary information using the enterprise knowledge base to generate features containing at least one production information feature;

[0119] The push module pushes the target push information based on the correlation analysis between the production information features and the demand information features.

[0120] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the aforementioned method for building a government and enterprise service platform using artificial intelligence algorithms.

[0121] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the aforementioned method for building a government and enterprise service platform using an artificial intelligence algorithm.

[0122] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0123] In this specification, the same or similar parts between the various embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the descriptions of the embodiments described later are relatively simple, and relevant parts can be referred to the descriptions of the foregoing embodiments.

[0124] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A government-enterprise service platform method constructed by an artificial intelligence algorithm, characterized in that, The method includes: Acquire enterprise video data and enterprise text data; The enterprise video data is processed to generate video summary information; The enterprise text data is parsed to generate an enterprise knowledge base that represents the enterprise's business priorities; Augmenting the video summary information with the enterprise knowledge base, generating a product identifier for in-process products Building a production information feature Fprod that contains the industry affiliation and the technical support of the product; Acquire external information data and extract demand information features from the external information data; Based on the correlation analysis between the production information features and the demand information features, target push information is provided; The enterprise text data includes at least enterprise patent information, industry qualification certificates, and management system certification certificates; The processing the enterprise video data to generate video summary information specifically comprises: identifying an in-production product identifier , calculating production efficiency corresponding to the in-production product identifier, and obtaining a production efficiency category based on preset efficiency comparison ; estimating an average value of the number of workers based on a pose estimation algorithm, and obtaining a human input rate category based on preset human threshold comparison , and integrating to obtain video summary information .

2. The method for constructing a government-enterprise service platform using artificial intelligence algorithms according to claim 1, characterized in that: Parsing the enterprise text data to generate an enterprise knowledge base includes constructing industry knowledge entries: It calls a pre-built authentication standard rule library, where each rule corresponds to an authentication standard and defines the inference type as either direct inference or semantic inference; The uploaded qualification certificates are extracted to identify the standard name of the certificate and the description text of the certification scope; and the certification standard rule base is queried based on the standard name, and industry knowledge entries are constructed by direct industry mapping or keyword extraction from the description text of the certification scope according to the inference type found in the query.

3. The method for constructing a government-enterprise service platform using artificial intelligence algorithms according to claim 2, characterized in that: Parsing the enterprise text data to generate an enterprise knowledge base also includes constructing technical knowledge entries: Call the patent classification database to convert the IPC or CPC classification numbers of patent documents into text descriptions of the classification numbers; The invention title, abstract, and classification number text description of the patent are concatenated into a text corpus; and a set of high-confidence technical keywords is selected by calculating the technical relevance score or TF-IDF weight score of candidate keywords in the text corpus in order to construct technical knowledge entries.

4. The method for constructing a government-enterprise service platform using artificial intelligence algorithms according to claim 3, characterized in that: The steps of expanding the video summary information using the enterprise knowledge base specifically include: semantically vectorizing the in-production product identifiers in the video summary information to generate product identifier feature vectors; traversing the industry knowledge entries in the enterprise knowledge base, calculating the similarity between the product identifier feature vectors and vectors representing industry semantics, and using the industry semantic vector with the highest similarity as the expanded information; and traversing the technical knowledge entries in the enterprise knowledge base, calculating the similarity between the product identifier feature vectors and vectors representing technical semantics, and using the technical semantic vector with the highest similarity as the expanded information.

5. The method for constructing a government-enterprise service platform using artificial intelligence algorithms according to claim 4, characterized in that: The correlation analysis based on the production information features and the demand information features specifically includes: extracting a set of keywords representing industry, product, technology, production efficiency demand, and human resource input demand from external information data; vectorizing the keyword sets for industry, product, and technology respectively, and comparing them with the corresponding vectors in the production information features; classifying the keyword sets for production efficiency demand and human resource input demand, and comparing them with the corresponding categories in the production information features; when at least four of the five dimensions of industry, product, technology, production efficiency, and human resource input are successfully matched, the target push information is determined.

6. A government-enterprise service platform system constructed using artificial intelligence algorithms, used to execute the government-enterprise service platform method constructed using artificial intelligence algorithms as described in claim 1, characterized in that, The system includes: Acquisition Module: The acquisition module obtains enterprise video data and enterprise text data; Summary information generation module: The summary information generation module processes the enterprise video data to generate video summary information; Summary information expansion module: The summary information expansion module uses the enterprise knowledge base to expand the video summary information and generate a feature containing at least one production information feature; Demand information feature extraction module: Expands the video summary information using the enterprise knowledge base to generate features containing at least one production information feature; The push module pushes the target push information based on the correlation analysis between the production information features and the demand information features.

7. An electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements a method for constructing a government and enterprise service platform using an artificial intelligence algorithm as described in any one of claims 1 to 5.

8. A computer-readable storage medium storing a computer program that, when executed by a processor, implements a method for constructing a government and enterprise service platform using an artificial intelligence algorithm as described in any one of claims 1 to 5.

Images